融合边缘保持与改进代价聚合的立体匹配算法

目的 立体匹配是计算机双目视觉的重要研究方向,主要分为全局匹配算法与局部匹配算法两类。传统的局部立体匹配算法计算复杂度低,可以满足实时性的需要,但是未能充分利用图像的边缘纹理信息,因此在非遮挡、视差不连续区域的匹配精度欠佳。为此,提出了融合边缘保持与改进代价聚合的立体匹配。方法 首先利用图像的边缘空间信息构建权重矩阵,与灰度差绝对值和梯度代价进行加权融合,形成新的代价计算方式,同时将边缘区域像素点的权重信息与引导滤波的正则化项相结合,并在多分辨率尺度的框架下进行代价聚合。所得结果经过视差计算,得到初始视差图,再通过左右一致性检测、加权中值滤波等视差优化步骤获得最终的视差图。结果 在Middlebury立体匹配平台上进行实验,结果表明,融合边缘权重信息对边缘处像素点的代价量进行了更加有效地区分,能够提升算法在各区域的匹配精度。其中,未加入视差优化步骤的21组扩展图像对的平均误匹配率较改进前减少3.48%,峰值信噪比提升3.57 dB,在标准4幅图中venus上经过视差优化后非遮挡区域的误匹配率仅为0.18%。结论 融合边缘保持的多尺度立体匹配算法有效提升了图像在边缘纹理处的匹配精度,进一步降低了非遮挡区域与视差不连续区域的误匹配率。     

特征融合的双目半全局匹配算法及其并行加速实现

目的 在微小飞行器系统中,如何实时获取场景信息是实现自主避障及导航的关键问题。本文提出了一种融合中心平均Census特征与绝对误差（AD）特征、基于纹理优化的半全局立体匹配算法（ADCC-TSGM）,并利用统一计算设备架构 （CUDA）进行并行加速。方法 使用沿极线方向的一维差分计算纹理信息,使用中心平均Census特征及AD特征进行代价计算,通过纹理优化的SGM算法聚合代价并获得初始视差图;然后,通过左右一致性检验检查剔除粗略视差图中的不稳定点和遮挡点,使用线性插值和中值滤波对视差图中的空洞进行填充;最后,利用GPU特性,对立体匹配中的代价计算、半全局匹配 （SGM）计算、视差计算等步骤使用共享内存、单指令多数据流 （SIMD）及混合流水线进行优化以提高运行速度。结果 在Quarter Video Graphics Array （QVGA）分辨率的middlebury双目图像测试集中,本文提出的ADCC-TSGM算法总坏点率较Semi-Global Block Matching （SGBM）算法降低36.1%,较SGM算法降低28.3%;平均错误率较SGBM算法降低44.5%,较SGM算法降低49.9%。GPU加速实验基于NVIDIA Jetson TK1嵌入式计算平台,在双目匹配性能不变的情况下,通过使用CUDA并行加速,可获得117倍以上加速比,即使相较于已进行SIMD及多核并行优化的SGBM,运行时间也减少了85%。在QVGA分辨率下,GPU加速后的运行帧率可达31.8 帧/s。结论 本文算法及其CUDA加速可为嵌入式平台提供一种实时获取高质量深度信息的有效途径,可作为微小飞行器、小型机器人等设备进行环境感知、视觉定位、地图构建的基础步骤。     

融合多特征表示和超像素优化的双目立体匹配

针对目前许多局部双目立体匹配方法在缺乏纹理区域、遮挡区域、深度不连续区域匹配精度低的问题,提出了基于多特征表示和超像素优化的立体匹配算法。通过在代价计算步骤中加入边缘信息特征,与图像局部信息代价相融合,增加了在视差计算时边缘区域的辨识度;在代价聚合步骤,基于超像素分割形成的超像素区域,利用米字骨架自适应搜索,得到聚合区域,对初始代价进行聚合;在视差精化步骤利用超像素分割信息,对匹配错误视差进行修正,提高匹配精度。基于Middlebury立体视觉数据集测试平台,与自适应权重AD-Census、FA等方法得出的视差图进行比较,该算法在深度不连续区域和缺乏纹理区域的匹配效果显著改善,提高了立体匹配精度。     

双目区域视差快速计算及测距算法

目的 双目测距对水面无人艇自主避障以及视觉侦察具有重要意义,但视觉传感器成像易受光照环境及运动模糊等因素的影响,基于经典Census变换的立体匹配代价计算方法耗时长,且视差获取精度差,影响测距精度。为了提高测距精度并保证算法运行速度,提出一种用于双目测距的快速立体匹配算法。方法 基于传统Census变换,提出一种新的比特串生成方法,在匹配点正方形支持窗口的各边等距各选3个像素点,共选出8个像素点,这8个像素点两两比较生成一个字节的比特串。将左右视场中的匹配点与待匹配点的比特串进行异或,得到两点的汉明距离,在各汉明距离中找到距离最小的像素点作为匹配像素点,两像素点的横坐标差为视差。本文采用区域视差计算的方法,在左右视场确定同一目标区域后进行视差提取和滤波,利用平均视差计算目标的距离。结果 本文算法与基于传统Census变换的立体匹配视差获取方法相比,在运算速度方面优势明显,时间稳定在0.4 s左右,用时仅为传统Census变换算法的1/5。在Middlebury数据集中的图像对teddy和cones上进行的算法运行时间对比实验中,本文基于Census变换改进的算法比已有的基于Census变换的匹配算法在运行时间上快了近20 s。在实际双目测距实验中,采用本文算法在1019 m范围内测距误差在5%以内,根据无人艇的运动特点和避障要求,通过分析可知该算法的测距精度可以满足低速无人艇的避障需求。结论 本文给出的基于改进Census变换的匹配算法在立体匹配速度上有大幅提高,提取目标视差用于测距,实际测距结果表明,本文算法能够满足水面无人艇的视觉避障要求。     

根据灰度值信息自适应窗口的半全局匹配

目的 立体匹配算法是立体视觉研究的关键点,算法的匹配精度和速度直接影响3维重建的效果。对于传统立体匹配算法来说,弱纹理区域、视差深度不连续区域和被遮挡区域的匹配精度依旧不理想,为此选择具有全局匹配算法和局部匹配算法部分优点、性能介于两种算法之间、且鲁棒性强的半全局立体匹配算法作为研究内容,提出自适应窗口与半全局立体匹配算法相结合的改进方向。方法 以通过AD（absolute difference）算法求匹配代价的半全局立体匹配算法为基础,首先改变算法匹配代价的计算方式,研究窗口大小对算法性能的影响,然后加入自适应窗口算法,研究自适应窗口对算法性能的影响,最后对改进算法进行算法性能评价与比较。结果 实验结果表明,匹配窗口的选择能够影响匹配算法性能、提高算法的适用范围,自适应窗口的加入能够提高算法匹配精度特别是深度不连续区域的匹配精度,并有效降低算法运行时间,对Cones测试图像集,改进的算法较改进前误匹配率在3个测试区域平均减少2.29%;对于所有测试图像集,算法运行时间较加入自适应窗口前平均减少28.5%。结论 加入自适应窗口的半全局立体匹配算法具有更优的算法性能,能够根据应用场景调节算法匹配精度和匹配速度。     

基于视差重映射的立体图像视觉舒适度提升

目的 针对人眼观看立体图像内容可能存在的视觉不舒适性,基于视差对立体图像视觉舒适度的影响,提出了一种结合全局线性和局部非线性视差重映射的立体图像视觉舒适度提升方法。方法 首先,考虑双目融合限制和视觉注意机制,分别结合空间频率和立体显著性因素提取立体图像的全局和局部视差统计特征,并利用支持向量回归构建客观的视觉舒适度预测模型作为控制视差重映射程度的约束;然后,通过构建的预测模型对输入的立体图像的视觉舒适性进行分析,就欠舒适的立体图像设计了一个两阶段的视差重映射策略,分别是视差范围的全局线性重映射和针对提取的潜在欠舒适区域内视差的局部非线性重映射;最后,根据重映射后的视差图绘制得到舒适度提升后的立体图像。结果 在IVY Lab立体图像舒适度测试库上的实验结果表明,相较于相关有代表性的视觉舒适度提升方法对于欠舒适立体图像的处理结果,所提出方法在保持整体场景立体感的同时,能更有效地提升立体图像的视觉舒适度。结论 所提出方法能够根据由不同的立体图像特征构建的视觉舒适度预测模型来自动实施全局线性和局部非线性视差重映射过程,达到既改善立体图像视觉舒适度、又尽量减少视差改变所导致的立体感削弱的目的,从而提升立体图像的整体3维体验。     

改进的跨尺度引导滤波立体匹配算法

作为双目三维重建中的关键步骤,双目立体匹配算法完成了从平面视觉到立体视觉的转化.但如何平衡双目立体匹配算法的运行速度和精度仍然是一个棘手的问题.本文针对现有的局部立体匹配算法在弱纹理、深度不连续等特定区域匹配精度低的问题,并同时考虑到算法实时性,提出了一种改进的跨多尺度引导滤波的立体匹配算法.首先融合AD和Census变换两种代价计算方法,然后采用基于跨尺度的引导滤波进行代价聚合,在进行视差计算时通过制定一个判断准则判断图像中每一个像素点的最小聚合代价对应的视差值是否可靠,当判断对应的视差值不可靠时,对像素点构建基于梯度相似性的自适应窗口,并基于自适应窗口修正该像素点对应的视差值.最后通过视差精化得到最终的视差图.在Middlebury测试平台上对标准立体图像对的实验结果表明,与传统基于引导滤波器的立体匹配算法相比具有更高的精度.     

基于平行双目立体视觉的测距系统

基于计算机视觉的理论基础,在PC机上搭建出了平行双目立体视觉测距系统的实验平台,在前人的基础上对传统的标定方法进行了改进,实现了双目摄像机的标定,并采用特征点匹配领域的热点与难点算法尺度不变特征变换(SIFT)特征匹配算法实现了立体匹配,恢复出物体的深度信息。通过实验与计算,最终验证了该方法理论分析的正确性及实践的可行性。     

双目机器视觉及RetinaNet模型的路侧行人感知定位

目的 行人感知是自动驾驶中必不可少的一项内容,是行车安全的保障。传统激光雷达和单目视觉组合的行人感知模式,设备硬件成本高且多源数据匹配易导致误差产生。对此,本文结合双目机器视觉技术与深度学习图像识别技术,实现对公共路权环境下路侧行人的自动感知与精准定位。方法 利用双目道路智能感知系统采集道路前景图像构建4种交通环境下的行人识别模型训练库;采用RetinaNet深度学习模型进行目标行人自动识别;通过半全局块匹配（semi-global block matching,SGBM）算法实现行人道路前景图像对的视差值计算;通过计算得出的视差图分别统计U-V方向的视差值,提出结合行人识别模型和U-V视差的测距算法,实现目标行人的坐标定位。结果 实验统计2.5 km连续测试路段的行人识别结果,对比人工统计结果,本文算法的召回率为96.27%。与YOLOv3（you only look once）和Tiny-YOLOv3方法在4种交通路况下进行比较,平均F值为96.42%,比YOLOv3和Tiny-YOLOv3分别提高0.9%和3.03%;同时,实验利用标定块在室内分别拍摄3 m、4 m和5 m不同距离的20对双目图像,验证测距算法,计算标准偏差皆小于0.01。结论 本文提出的结合RetinaNet目标识别模型与改进U-V视差算法能够实现对道路行人的检测,可以为自动驾驶的安全保障提供技术支持,具有一定的应用价值。     

有约束Patch-Match框架下的非刚体匹配算法

目的 非刚性物体进行匹配时,往往需要对图像中存在的非刚性形变目标进行快速精确的配准,进而实现对图像的后续处理和分析,实现快速而准确的非刚体匹配显得尤为重要。针对传统特征点匹配方法在非刚性物体匹配中准确性差的问题,本文提出了一种基于DAISY算子和有约束Patch-Match的非刚体密集匹配算法。方法 首先对参考图像和待匹配图像生成DAISY特征描述子,其次对两幅图像进行超像素分割,形成相互邻接但没有重叠的超像素块结构,并以其为单元,计算初始位置上对应每一个像素的DAISY特征算子聚合代价。然后,采用Patch-Match算法对整幅图像进行传播和变异,在变异过程中,通过图像预处理和分析得到的先验知识对位置标签的变异窗口进行局部空间约束,使得每个像素的位置标签在该空间范围内随机更新,计算新的聚合代价,保留代价较小的位置标签,重复迭代此过程,直到聚合代价不发生变化或者达到最大迭代次数为止。结果 实验选取了标准数据集、10幅分别由TFDS（the trucking fault dynamic image detection system）线阵列相机和框幅式相机采集的包含非刚体的图像进行匹配,均取得了较好的匹配效果,经验证,本文方法的匹配精度为86%,误匹配点的平均匹配误差为5个像素左右,是传统基于SIFT特征光流匹配方法误差的一半,并且本文采用的DAISY算子在特征提取速度上是Dense SIFT（dense scale invariant feature transform）特征提取算法的2~3倍,大大提升了图像匹配的效率。结论 本文提出了一种非刚体密集匹配算法,针对非刚体变化的不确定性采用密集特征点进行最优化搜索匹配。本文算法对包含小范围非刚性变化的图像匹配上具有较好的适应性,且匹配精度高,视觉效果好,鲁棒性强。     

Accurate real-time stereo correspondence using intra- and inter-scanline optimization

This paper deals with a novel stereo algorithm that can generate accurate dense disparity maps in real time. The algorithm employs an effective cross-based variable support aggregation strategy within a scanline optimization framework. Rather than matching intensities directly, the use of adaptive support aggregation allows for precisely handling the weak textured regions as well as depth discontinuities. To improve the disparity results with global reasoning, we reformulate the energy function on a tree structure over the whole 2D image area, as opposed to dynamic programming of individual scanlines. By applying both intra- and inter-scanline optimizations, the algorithm reduces the typical ’streaking’ artifact while maintaining high computational efficiency. The experimental results are evaluated on the Middlebury stereo dataset, showing that our approach is among the best for all real-time approaches. We implement the algorithm on a commodity graphics card with CUDA architecture, running at about 35 fames/s for a typical stereo pair with a resolution of 384×288 and 16 disparity levels.     

Genetic programming on graphics processing units

The availability of low cost powerful parallel graphics cards has stimulated the port of Genetic Programming (GP) on Graphics Processing Units (GPUs). Our work focuses on the possibilities offered by Nvidia G80 GPUs when programmed in the CUDA language. In a first work we have showed that this setup allows to develop fine grain parallelization schemes to evaluate several GP programs in parallel, while obtaining speedups for usual training sets and program sizes. Here we present another parallelization scheme and optimizations about program representation and use of GPU fast memory. This increases the computation speed about three times faster, up to 4 billion GP operations per second. The code has been developed within the well known ECJ library and is open source.     

A Performance Study on Different Cost Aggregation Approaches Used in Real-Time Stereo Matching

Many vision applications require high-accuracy dense disparity maps in real-time and online. Due to time constraint, most real-time stereo applications rely on local winner-takes-all optimization in the disparity computation process. These local approaches are generally outperformed by offline global optimization based algorithms. However, recent research shows that, through carefully selecting and aggregating the matching costs of neighboring pixels, the disparity maps produced by a local approach can be more accurate than those generated by many global optimization techniques. We are therefore motivated to investigate whether these cost aggregation approaches can be adopted in real-time stereo applications and, if so, how well they perform under the real-time constraint. The evaluation is conducted on a real-time stereo platform, which utilizes the processing power of programmable graphics hardware. Six recent cost aggregation approaches are implemented and optimized for graphics hardware so that real-time speed can be achieved. The performances of these aggregation approaches in terms of both processing speed and result quality are reported.     

基于可变窗口视差优化的并行立体匹配

针对现有局部立体匹配算法在计算精度和执行效率之间的权衡问题,提出一种基于可变窗口视差优化的并行立体匹配方法。为弥补ESAW（Exponential Step Adaptive Weight）代价聚合方法所造成的精度损失,在视差优化阶段,为每个误差点建立一个基于颜色相似度和欧式距离的可变窗口,并将误差点分为遮挡和误匹配两类,针对不同的类型采用不同的优化策略;利用CUDA（Compute Unified Device Architecture）技术将算法在图形处理器上进行并行优化和验证。实验结果表明,与现有Middlebury测试平台中列出的并行立体匹配算法相比,具有更好的计算精度。     

GPU-accelerated string matching for database applications

Implementations of relational operators on GPU processors have resulted in order of magnitude speedups compared to their multicore CPU counterparts. Here we focus on the efficient implementation of string matching operators common in SQL queries. Due to different architectural features the optimal algorithm for CPUs might be suboptimal for GPUs. GPUs achieve high memory bandwidth by running thousands of threads, so it is not feasible to keep the working set of all threads in the cache in a naive implementation. In GPUs the unit of execution is a group of threads and in the presence of loops and branches, threads in a group have to follow the same execution path; if some threads diverge, then different paths are serialized. We study the cache memory efficiency of single- and multi-pattern string matching algorithms for conventional and pivoted string layouts in the GPU memory. We evaluate the memory efficiency in terms of memory access pattern and achieved memory bandwidth for different parallelization methods. To reduce thread divergence, we split string matching into multiple steps. We evaluate the different matching algorithms in terms of average- and worst-case performance and compare them against state-of-the-art CPU and GPU libraries. Our experimental evaluation shows that thread and memory efficiency affect performance significantly and that our proposed methods outperform previous CPU and GPU algorithms in terms of raw performance and power efficiency. The Knuth–Morris–Pratt algorithm is a good choice for GPUs because its regular memory access pattern makes it amenable to several GPU optimizations.     

Massively parallel GPU computing for fast stereo correspondence algorithms

Current accurate stereo matching algorithms employ some key techniques that are not suitable for parallel GPU architecture. It will be tricky and cumbersome to directly take these techniques into GPU applications. Trying to tackle this difficulty, we design two GPU-based stereo matching algorithms, one using a local fixed aggregation window whose size is configurable, and the other using an adaptive aggregation window which only includes necessary pixels. We use the winner-takes-all (WTA) principle for optimization and a plain voting refinement for post-processing; both do not need complex data structures. We aim to implement on GPU platforms fast stereo matching algorithms that produce results with same-level quality as other WTA local dense methods that use window-based cost aggregation. In our GPU-based implementation of the fixed window partially demosaiced CFA stereo matching application, accelerations up to 20 times are obtained for large size images. In our GPU-based implementation of the adaptive window color stereo matching application, experiment results show that it can handle four pairs of standard images from Middlebury database within roughly 100 ms.     

Stereo correspondence using efficient hierarchical belief propagation

In this paper, a new algorithm is presented to compute the disparity map from a stereo pair of images by using Belief Propagation (BP). While many algorithms have been proposed in recent years, the real-time computation of an accurate disparity map is still a challenging task. The computation time and run-time memory requirements are two very important factors for all real-time applications. The proposed algorithm divides the matching process into two steps; they are initial matching and disparity map refinement. Initial matching is performed by memory efficient hierarchical belief propagation algorithm that uses less than half memory at run-time and minimizes the energy function at much faster rate as compare to other hierarchical BP algorithms that makes it more suitable for real-time applications. Disparity map refinement uses a simple but very effective single-pass approach that improves the accuracy without affecting the computation cost. Experiments by using Middlebury dataset demonstrate that the performance of our algorithm is the best among other real-time stereo matching algorithms.     

Efficient magnetohydrodynamic simulations on distributed multi-GPU systems using a novel GPU Direct–MPI hybrid approach

Modern graphics processing units (GPUs) have been widely utilized in magnetohydrodynamic (MHD) simulations in recent years. Due to the limited memory of a single GPU, distributed multi-GPU systems are needed to be explored for large-scale MHD simulations. However, the data transfer between GPUs bottlenecks the efficiency of the simulations on such systems. In this paper we propose a novel GPU Direct–MPI hybrid approach to address this problem for overall performance enhancement. Our approach consists of two strategies: (1) We exploit GPU Direct 2.0 to speedup the data transfers between multiple GPUs in a single node and reduce the total number of message passing interface (MPI) communications; (2) We design Compute Unified Device Architecture (CUDA) kernels instead of using memory copy to speedup the fragmented data exchange in the three-dimensional (3D) decomposition. 3D decomposition is usually not preferable for distributed multi-GPU systems due to its low efficiency of the fragmented data exchange. Our approach has made a breakthrough to make 3D decomposition available on distributed multi-GPU systems. As a result, it can reduce the memory usage and computation time of each partition of the computational domain. Experiment results show twice the FLOPS comparing to common 2D decomposition MPI-only implementation method. The proposed approach has been developed in an efficient implementation for MHD simulations on distributed multi-GPU systems, called MGPU–MHD code. The code realizes the GPU parallelization of a total variation diminishing (TVD) algorithm for solving the multidimensional ideal MHD equations, extending our work from single GPU computation (Wong et al., 2011) to multiple GPUs. Numerical tests and performance measurements are conducted on the TSUBAME 2.0 supercomputer at the Tokyo Institute of Technology. Our code achieves 2 TFLOPS in double precision for the problem with 1200³ grid points using 216 GPUs.     

基于改进Census变换的多特性立体匹配算法

针对当前立体匹配算法存在的匹配准确率低，难以达到实用的高精度水平的问题，提出了一种基于改良的Census变换与色彩信息和梯度测度相结合的多特性立体匹配算法，实现高精度的双目立体匹配。算法首先在初始代价匹配阶段，将改进的Census变换、色彩和梯度测度赋权求和得出可靠的初始匹配代价；在聚合阶段，采取高效快捷的最小生成树聚合，获得匹配代价矩阵；最后根据胜者为王法则得到初始视差图，并引入左右一致性检测等策略优化视差图，获得高精度的视差图，实验阶段对源自Middlebury上的标准测试图进行测试验证，实验结果表明，经本文算法处理得到的15组测试数据集的视差图在非遮挡区域的平均误匹配率为6.81%，算法实时响应性优良。